Attempts to improve performance. Not much gain yet, but good to experiment what has effect and what hasn't.

Added 'DO_BENCHMARK_PYRAMID' to CcdPhysicsDemo.

src/BulletCollision/CollisionDispatch/btSimulationIslandManager.cpp

@@ -122,12 +122,9 @@ class btPersistentManifoldSortPredicate
 {
 	public:
 
-		bool operator() ( const btPersistentManifold* lhs, const btPersistentManifold* rhs )
+		SIMD_FORCE_INLINE bool operator() ( const btPersistentManifold* lhs, const btPersistentManifold* rhs )
 		{
-			int rIslandId0,lIslandId0;
-			rIslandId0 = getIslandId(rhs);
-			lIslandId0 = getIslandId(lhs);
-			return lIslandId0 < rIslandId0;
+			return getIslandId(lhs) < getIslandId(rhs);
 		}
 };
 
@@ -142,8 +139,17 @@ void btSimulationIslandManager::buildAndProcessIslands(btDispatcher* dispatcher,
 {
 
 	
-	BEGIN_PROFILE("islandUnionFindAndHeapSort");
+	
+	if (0)
+	{
+		int maxNumManifolds = dispatcher->getNumManifolds();
+		btCollisionDispatcher* colDis = (btCollisionDispatcher*)dispatcher;
+		btPersistentManifold** manifold = colDis->getInternalManifoldPointer();
+		callback->ProcessIsland(&collisionObjects[0],collisionObjects.size(),manifold,maxNumManifolds, 0);
+		return;
+	}
 
+	BEGIN_PROFILE("islandUnionFindAndHeapSort");
 	
 	//we are going to sort the unionfind array, and store the element id in the size
 	//afterwards, we clean unionfind, to make sure no-one uses it anymore
@@ -152,9 +158,11 @@ void btSimulationIslandManager::buildAndProcessIslands(btDispatcher* dispatcher,
 	int numElem = getUnionFind().getNumElements();
 
 	int endIslandIndex=1;
+	int startIslandIndex;
+
 
 	//update the sleeping state for bodies, if all are sleeping
-	for (int startIslandIndex=0;startIslandIndex<numElem;startIslandIndex = endIslandIndex)
+	for ( startIslandIndex=0;startIslandIndex<numElem;startIslandIndex = endIslandIndex)
 	{
 		int islandId = getUnionFind().getElement(startIslandIndex).m_id;
 		for (endIslandIndex = startIslandIndex+1;(endIslandIndex<numElem) && (getUnionFind().getElement(endIslandIndex).m_id == islandId);endIslandIndex++)

src/BulletCollision/CollisionDispatch/btUnionFind.h

@@ -18,6 +18,8 @@ subject to the following restrictions:
 
 #include "../../LinearMath/btAlignedObjectArray.h"
 
+	#define USE_PATH_COMPRESSION 1
+
 struct	btElement
 {
 	int	m_id;
@@ -79,6 +81,7 @@ class btUnionFind
 			if (i == j) 
 				return;
 
+#ifndef USE_PATH_COMPRESSION
 			//weighted quick union, this keeps the 'trees' balanced, and keeps performance of unite O( log(n) )
 			if (m_elements[i].m_sz < m_elements[j].m_sz)
 			{ 
@@ -88,6 +91,9 @@ class btUnionFind
 			{ 
 				m_elements[j].m_id = i; m_elements[i].m_sz += m_elements[j].m_sz; 
 			}
+#else
+			m_elements[i].m_id = j; m_elements[j].m_sz += m_elements[i].m_sz; 
+#endif //USE_PATH_COMPRESSION
 		}
 
 		int find(int x)
@@ -98,7 +104,7 @@ class btUnionFind
 			while (x != m_elements[x].m_id) 
 			{
 		//not really a reason not to use path compression, and it flattens the trees/improves find performance dramatically
-		#define USE_PATH_COMPRESSION 1
+	
 		#ifdef USE_PATH_COMPRESSION
 				//
 				m_elements[x].m_id = m_elements[m_elements[x].m_id].m_id;

src/BulletCollision/CollisionShapes/btBoxShape.cpp

@@ -49,10 +49,9 @@ void	btBoxShape::calculateLocalInertia(btScalar mass,btVector3& inertia)
 	btScalar ly=btScalar(2.)*(halfExtents.y());
 	btScalar lz=btScalar(2.)*(halfExtents.z());
 
-	inertia[0] = mass/(btScalar(12.0)) * (ly*ly + lz*lz);
-	inertia[1] = mass/(btScalar(12.0)) * (lx*lx + lz*lz);
-	inertia[2] = mass/(btScalar(12.0)) * (lx*lx + ly*ly);
-
+	inertia.setValue(mass/(btScalar(12.0)) * (ly*ly + lz*lz),
+					mass/(btScalar(12.0)) * (lx*lx + lz*lz),
+					mass/(btScalar(12.0)) * (lx*lx + ly*ly));
 
 }
 

src/BulletCollision/CollisionShapes/btSphereShape.cpp

@@ -68,7 +68,7 @@ void btSphereShape::getAabb(const btTransform& t,btVector3& aabbMin,btVector3& a
 void	btSphereShape::calculateLocalInertia(btScalar mass,btVector3& inertia)
 {
 	btScalar elem = btScalar(0.4) * mass * getMargin()*getMargin();
-	inertia[0] = inertia[1] = inertia[2] = elem;
+	inertia.setValue(elem,elem,elem);
 
 }